Why nobody chose your AI stack
Most teams did not choose their AI stack. It accumulated: someone expensed ChatGPT, engineering got Copilot with their Microsoft license, marketing signed up for Claude, and Gemini arrived with Workspace. Nobody decided this, which is why nobody owns it.
Only 40% of companies had purchased an official LLM subscription — but workers at more than 90% of companies surveyed reported using personal AI tools for work.
That gap is the whole problem. The tools people actually use hold company data, sit outside any policy, and disappear when the employee does. Meanwhile the subscription finance approved goes half-used.
It is worth being precise about what kind of problem this is. It is not a discipline problem, and banning the personal tools does not fix it — people reached for them because the approved tool did not do the thing they needed at the moment they needed it. Consolidation only works if the thing you consolidate onto is genuinely broader than what it replaces.
The 20-minute audit
You do not need a procurement exercise to get a usable picture. Four passes, and none of them requires anyone's permission:
- Pull the card statements for the last three months. Search for the obvious vendor names, then search again for the ones that bill through an app store or a reseller. Expensed subscriptions are the ones that never show up in a SaaS inventory tool.
- List every AI feature that arrived inside something you already pay for. Copilot inside Microsoft 365, Gemini inside Workspace, the assistant inside your CRM. These are not free; they are bundled, which is different, and they are often the reason a paid seat elsewhere goes unused.
- Ask three people what they actually open. Not in a survey — in a hallway. The answer is usually one or two tools they pay for personally, and the reason is almost always a specific capability rather than brand preference.
- Count the seats that logged in this month. Every vendor exposes this somewhere. The ratio of paid seats to active seats is the single number that decides whether consolidation saves money or only tidies the bill.
Write the result as one table: tool, monthly cost, paid seats, active seats, and what it does that nothing else on the list does. That last column is the one that decides everything, and it is the one people skip.
What consolidation actually saves
Three things, in descending order of how reliably they show up. Be suspicious of anyone who promises them in a different order.
1. Unused seats, immediately
This is the real, bankable saving, and it has nothing to do with which platform you pick. If you are paying for 40 seats and 22 people signed in last month, you were going to save that money by counting. Consolidation just makes the count happen.
2. Duplicated capability
Two tools that both summarize documents are one tool and one line item. This saving is real but smaller than it looks, because the second tool usually does one thing the first does not — and the person who signed up for it knows exactly what that thing is.
3. The cost of not knowing
Unmeasurable up front, and the largest of the three in a bad month. Personal accounts holding company documents cannot be audited, cannot be revoked when someone leaves, and cannot be excluded from a vendor's training by policy you do not control. You cannot put a number on this in advance, so do not let anyone sell it to you as one.
ChatGPT's share of consumer AI web traffic fell from 76% to 53% in eleven months, while Gemini rose from under 9% to roughly 28%.
The reason that matters for a consolidation decision is not which vendor is winning. It is that the answer changed twice in a year — so committing your team to a single lab's models is a bet on a leaderboard that keeps reordering itself.
When staying split is the right call
Consolidating is not automatically correct. Four situations where it is not:
- Under about five people. The coordination cost of a shared workspace exceeds the saving. Expense the individual subscriptions and revisit at ten.
- One workflow is deeply embedded in one vendor. If your engineers live inside a coding assistant in the editor, that is not a chat tool and it does not consolidate away. Treat it as a development tool and leave it alone.
- The bundled seat is genuinely free. If an assistant comes with a license you already pay for and it covers the use case, moving it costs money rather than saving it.
- You are mid-renewal on a long contract. Consolidate at the renewal date, not before. Paying twice during an overlap wipes out a year of the saving.
If none of those apply and your active-seat ratio is under about two thirds, consolidation is straightforwardly worth doing. If they do apply, the audit was still worth twenty minutes: you now know what you are paying for, which is more than most teams can say.
Akili is one way to run a consolidated workspace — nine model providers in one conversation, one pooled bill, and answers grounded in your own documents. That is the product, and it is the last time this page mentions it. If the audit says stay split, stay split. Prices are published either way.